Sustainable AI: 3 tools to measure the environmental impact of ML solutions

Sustainable AI: 3 tools to measure the environmental impact of ML solutions

Richard Brown

22 May 2023 - 7 min read

AIMachine LearningSustainability
Sustainable AI: 3 tools to measure the environmental impact of ML solutions

Artificial intelligence (AI) has become an essential tool for organisations across a wide range of industries, from healthcare to agriculture. However, the development and deployment of the machine learning (ML) models that power these solutions are not without environmental consequences.

The high computational power required for model training and inference, as well as the associated energy consumption, can result in a significant carbon footprint with varying levels of carbon intensity. Research has shown that training a single large language model can emit as much as 626,155 pounds of carbon dioxide equivalent (CO2e), which is roughly equal to the total lifetime carbon footprint of five cars.

In this article, we will firstly explain how AI is impacting the environment, before then exploring three tools that can help measure the carbon footprint of your ML models.

How AI is impacting the environment

An AI carbon footprint is the amount of carbon dioxide (CO2) and other greenhouse gases emitted during the production, training and use of AI systems. These systems are becoming increasingly complex and require more data to train, which means that their carbon footprint is growing.

Google has shown that AI contributes to 10-15% of their overall electricity usage, which was 18.3 terawatt hours in 2021. That would mean that Google’s AI burns around 2.3 terawatt hours annually, nearly double the amount of electricity required to power the London underground annually.

The majority of the AI carbon footprint comes from the energy used to train and run AI systems. AI systems are typically trained on large datasets that are stored in data centres. The greater the amount of energy that these data centres consume, the greater the model’s carbon footprint.

The following tools provide organisations with ways to measure the carbon footprint of their machine learning models and AI systems. By using these tools, organisations can gain insights into the environmental impact of their AI systems and take action to reduce their carbon emissions.

Microsoft Sustainability Calculator

As more enterprise organisations turn to cloud services like Azure Cognitive Service and Azure OpenAI Service for their AI requirements, it becomes increasingly important to evaluate the carbon footprint of these services. The Microsoft Sustainability Calculator is available as a Power BI application and is designed to help organisations measure and track the carbon footprint of their IT infrastructure.

The calculator provides a comprehensive view of the environmental impact of using cloud services offered by Azure, enabling organisations to make informed decisions about their AI infrastructure and reduce their overall carbon footprint.

The application includes a variety of data sources, including industry-standard emission factors and energy consumption metrics, to provide accurate estimates of carbon emissions and energy usage. It also offers a range of reporting options, giving users the ability to create custom reports that show the carbon impact of their technology investments over time.

Three key benefits of the calculator include:

  1. Identifies areas for improvement: The Microsoft Sustainability Calculator can help organisations identify areas for improvement in their IT infrastructure. By identifying energy-intensive areas, organisations can take steps to reduce their carbon footprint and become more sustainable.
  2. Tracks progress over time: The calculator also allows organisations to track their progress over time. By setting baseline measurements and comparing them to current measurements, organisations can monitor their progress towards sustainability goals.
  3. Enterprise support: With the Microsoft Sustainability Calculator, organisations have access to dedicated enterprise support offered by Microsoft. This support includes access to technical experts and customised reporting capabilities to help organisations track their carbon emissions over time. This level of support can help ensure that organisations are accurately measuring and reducing their carbon footprint, as well as aligning their sustainability goals with their broader business strategy.

Organisations should, nevertheless, keep in mind that the Microsoft Sustainability Calculator provides only an estimate of the carbon footprint of your machine learning model. The actual carbon footprint may vary depending on a variety of factors, including the specifics of your local power supply and any optimisations you have made to reduce emissions during training.

ML CO2 Impact Calculator

The ML CO2 Impact calculator is a useful tool for measuring the equivalent carbon dioxide generated during the training of machine learning models. Available online, users receive an estimate about the carbon impact of their solutions by inputting information about hardware, runtime and cloud provider.

Upon completion of calculations, the calculator produces two figures: the raw carbon emissions and the estimated offset carbon emissions, which may be subject to change depending on the cloud provider's grid.

Three key benefits to the calculator include:

  1. Availability: The calculator is a free tool that is available online. This means that it can be accessed by anyone with an internet connection, making it a convenient option for organisations of all sizes.
  2. Compatibility: The calculator is compatible with a wide range of cloud services, including Microsoft Azure and Amazon Web Services. This allows organisations that use cloud-based machine learning services to easily estimate their carbon footprint and take steps to reduce their environmental impact.
  3. Active community: Led by researchers and academics, the calculator is based on the latest research on the carbon emissions of machine learning models.

One potential drawback of using the ML CO2 Impact calculator is that it may not provide an accurate estimate for all scenarios. The calculator relies on assumptions and estimations for certain factors, such as the energy efficiency of the hardware and the carbon intensity of the electricity grid used by the cloud provider. It’s important to consider this limitation and to use the estimate provided by the calculator as a rough guide rather than a precise measurement.


CodeCarbon is a Python library that can be used to measure the carbon emissions of machine learning training and inference workloads. It can be used in a variety of settings including local development environments, cloud computing instances and high-performance computing clusters.

Three benefits of CodeCarbon are:

  1. Granular measurement: CodeCarbon provides a fine-grained measurement of carbon emissions by tracking the energy consumption of individual CPU and GPU devices during the training and inference process. This level of detail enables developers to identify which components of the system are contributing the most to carbon emissions and take action to optimise them.
  2. Easy integration: CodeCarbon is easy to integrate into existing machine learning workflows since it is built as a Python library. It can be used with popular machine learning frameworks such as TensorFlow, PyTorch, and scikit-learn.
  3. Open-source: CodeCarbon is an open-source project, which means it is freely available for use and can be modified to meet the specific needs of individual organisations. This can help organisations save on licensing fees while also contributing to a larger community effort to reduce the environmental impact of machine learning.

One potential drawback of using CodeCarbon is that it requires some technical expertise and knowledge of traditional machine learning development, particularly with Python. This can be a barrier for organisations that do not have in-house expertise or may be new to machine learning.


Measuring the carbon footprint of your machine learning model is a crucial step towards building more sustainable and environmentally friendly AI solutions. The three tools discussed in this article, Microsoft Sustainability Calculator, ML CO2 Impact Calculator and CodeCarbon, offer organisations different ways to estimate their carbon emissions and identify areas for improvement.

While these tools provide valuable insights, it's important to keep in mind that they are estimates and may not provide accurate measurements in all scenarios. Nevertheless, they can be used as a starting point for organisations to reduce their carbon footprint and move towards a more sustainable future.

Audacia is a leading software development company based in the UK, headquartered in Leeds. We have experience delivering artificial intelligence solutions across industries such as manufacturing, rail and healthcare.

To find out more about the artificial intelligence solutions we offer, contact us today on 0113 543 1300 or at

Ebook Available

How to maximise the performance of your existing systems

Free download

Richard Brown is the Technical Director at Audacia, where he is responsible for steering the technical direction of the company and maintaining standards across development and testing.